26.07.2013 Views

Full Report - Center for Collaborative Education

Full Report - Center for Collaborative Education

Full Report - Center for Collaborative Education

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

salient trends across time. Appropriate statistical<br />

tests – chi-square, t-test, or Fisher’s Exact test, with<br />

α =.05 <strong>for</strong> all – were used to determine the significance<br />

of the differences in outcomes among populations<br />

and among LEP students enrolled in schools<br />

showing different characteristics and in different<br />

types of ELL programs. Effect size was calculated<br />

where appropriate.<br />

As noted earlier in this Appendix, the dropout<br />

analysis conducted <strong>for</strong> this report was done in the<br />

separate dropout database. Annual dropout rates<br />

were determined <strong>for</strong> students enrolled in middle<br />

school and high school grades. 3 For summer<br />

dropouts, behavioral, academic (namely MEPA and<br />

MCAS), ELL program and school characteristics<br />

data from the prior year (the last school year completed)<br />

were assigned to the student. For instance,<br />

students who dropped out during the summer of<br />

2008 were assigned their SY2008 values <strong>for</strong> these<br />

SY2009 variables. Basic frequencies and cross-tabulations<br />

were conducted and statistical significance<br />

was determined by running chi-square 4 tests<br />

(α =.05) and by determining the effect size.<br />

In addition, an in-depth analysis was conducted to<br />

explore the impact of student-level characteristics<br />

and school environments on individual achievement<br />

as measured by MCAS per<strong>for</strong>mance in the English<br />

Language Arts and Math. 5 We identified hierarchical<br />

linear modeling (HLM) as the preferred method<br />

of analysis; due to the similarity of educational<br />

experiences between students in a particular school,<br />

traditional multiple regression techniques would<br />

underestimate the correlation between school-level<br />

variables and there<strong>for</strong>e the standard error, likely<br />

resulting in spuriously significant relationships. Variables<br />

of interest were identified through a review<br />

of the literature, the descriptive analyses, and in<br />

consultation with OELL.<br />

Six two-level models were tested: MCAS ELA<br />

per<strong>for</strong>mance <strong>for</strong> LEPs in SY2009 at elementary,<br />

middle, and high school levels and MCAS Math<br />

per<strong>for</strong>mance <strong>for</strong> LEPs in SY2009 at elementary,<br />

middle, and high school levels. For the MCAS ELA<br />

models, elementary grades included grades 4-5,<br />

middle school grades included Grades 6-8, and high<br />

school grades included Grades 9-12. For MCAS<br />

Math models, elementary grades included Grades<br />

3-5, middle school grades included Grades 6-8, and<br />

high school grades included Grades 9-12.<br />

Be<strong>for</strong>e including all explanatory variables in the<br />

models, the intraclass correlation coefficient (ICC)<br />

was calculated to verify that a hierarchical model<br />

was needed (see Table 47). Next, we checked<br />

multicollinearity to determine the model with valid<br />

significance levels. Usually, higher correlations<br />

among independent variables will result in a higher<br />

condition index, and a variable may have to be removed<br />

from the model <strong>for</strong> accurate estimation with<br />

significance testing. Within the set of student-level<br />

variables, Attendance Rate and Mobility were highly<br />

correlated at the elementary and middle school<br />

levels. The condition index was also high. Mobility<br />

was removed from the model and Attendance<br />

Rate was retained because the attendance variable<br />

structure (ratio rather than categorical) provides<br />

the opportunity <strong>for</strong> more detailed analysis. Percent<br />

Mobility, a school-level variable representing the<br />

percentage of the student population that changes<br />

schools between October and June of a given<br />

school year, was found to be strongly associated<br />

with LEP Density. Percent Mobility was removed<br />

from the model because LEP Density was considered<br />

of more interest to this analysis. Finally, Highly<br />

Qualified Teachers, a school-level variable representing<br />

the percentage of the teaching staff that is<br />

considered highly qualified, was also removed from<br />

the analysis, because the structure of the variable<br />

biased the analysis towards schools with highly<br />

qualified teachers on staff.<br />

118 Improving <strong>Education</strong>al Outcomes of English Language Learners in Schools and Programs in Boston Public Schools

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!